vLLM

Hugging Face's production-ready toolkit for deploying LLMs. Optimized for high throughput with tensor parallelism, quantization, and Flash Attention.

Free, open source alternative to OpenAI API. Run LLMs, generate images, audio, and more locally or on-prem with no GPU required.

LLM inference in pure C/C++. Run LLaMA and other models on consumer hardware with CPU and GPU support. The engine behind many local AI apps.